A process of inspecting a shape of a circuit pattern formed on a semiconductor wafer, for foreign matter and the like in a process of manufacturing a semiconductor device plays an important role in improving yields and establishing reliability. In the aforementioned process of inspecting, the inspecting is performed mainly by an inspection system utilizing light.
In particular, as semiconductor devices have become finer in recent years, it is difficult to observe the details of a defect only by the aforementioned inspecting. Accordingly, a defect review system, which magnifies and observes the defect found in the inspecting, is important. The defect review system is an apparatus for acquiring an image of a region including each defect on the basis of position information of the defect on a semiconductor wafer detected by another inspection system, such as a defect inspection system.
The defect review system is required to observe a defect with a size equal to or smaller than that of a process node. Accordingly, a charged particle microscope, which utilizes charged particle beams such as electron beams and provides nanometer-level resolution, is mainly used. However, the defect review system employing the charged particle microscope has a throughput lower than that of an inspection system utilizing light. Thus, it is important to increase the throughput of the defect review system for the sake of improving inspection efficiency.
One of the factors reducing the throughput may be coordinate accuracy of an observation position. The defect review system employing the charged particle microscope acquires an image at a defect position and an image (reference image) to be acquired if the defect does not exist at low magnification with a wide field of view for the sake of identifying the defect position, compares the images with each other and identifies the defect position, and subsequently acquires a defect image for observation at high magnification where the defect is enlarged and displayed.
Here, referring to FIG. 1, an operation of a typical defect review system is briefly described. First, based on an image acquisition preparatory sequence 121 and setting of an image acquisition condition setting 122, a sequence 123 of acquiring a reference image for search and a sequence 124 of acquiring a defect image for search are performed. Then, according to a defect position extraction 125, a sequence 126 of acquiring a defect image for observation acquires a defect image at an appropriate position. Subsequently, the sequences 123, 124 and 125 are performed until the defect review is completed. An acquisition condition used in the image acquisition condition setting 122 has been preset in a recipe to be read in the image acquisition preparatory sequence 121. The setting condition of an acquisition image is not changed in the middle of review.
In a case of a poor coordinate accuracy, a defect cannot accurately be included in a narrow and high magnification field of view. In order to avoid that, it is required to search a relatively wide field of view at low magnification for a defect. Accordingly, if acquisition of an image for search is set, the acquisition of an image for search is performed even on a defect that is originally acquirable in a narrow field of view with high magnification. This leads to reduction in throughput. Accordingly, it is important to improve the coordinate accuracy.
Among techniques improving coordinate accuracy and throughput, for example in Patent Literature 1, a technique is disclosed that absorbs a coordinate error by position alignment between design data of a semiconductor integrated circuit and an image at a reference position, and thereby acquires an image for observation without acquiring any image for search and improves the throughput. Further, in Patent Literature 2, a technique is disclosed that removes a coordinate error owing to a coordinate offset between system, subsequently calculates variation in inspection by statistical processing and determines a magnification of an image for search using calculated data.